ACL2026
DataSeer: A Manager-Centric Collaborative Multi-Agent Framework with Multi-Branch Reasoning for Automated Insight Discovery
Suchen Liu, Yuanfeng Song, Jun Gao, Xing Chen
Abstract
The increasing complexity of data drives the demand for automated insight discovery. While LLMs and agent technologies have advanced data analysis, existing methods struggle with maintaining contextual coherence, achieving sufficient coverage (due to single-path exploration), and adapting rigid planning to dynamic data discovery. We propose DataSeer, a collaborative multi-agent framework for automated insight discovery. Our first contribution is a Manager-Centric Collaborative Framework, where the Manager ensures cross-episode contextual coherence through a dual-layer memory system with compression, consolidation, and retrieval, alongside dynamic prompt editing, coordinating the overall process between the Planner and Executor. Second, we optimize the planning and execution components: the Planner employs multi-role discussion for adaptive sub-goal generation and plan refinement; the Executor is endowed with tactical autonomy for exploratory execution and incorporates realtime multi-dimensional self-assessment to guarantee insight quality. Third, we design Multi-Branch Reasoning that executes multiple discovery trajectories and synthesizes outcomes through LLM-based aggregation, improving coverage and reducing single-path stochasticity. Experiments on InsightBench and InsightEval show that DataSeer outperforms baselines, achieving improvements of 18.7% and 12.1% in insight-level scores, and 11.6% and 10.3% in summary-level scores, respectively. * Corresponding authors. * Be specific and actionable * Focus on a single , well -defined task * Consider the available data and schema * Build upon previous steps in the history When creating your plan , consider : * What specific question needs to be answered next ? * What data exploration or analysis is needed ? * If the history contains results from previous plans , evaluate these results and adjust your plan accordingly ### Output Format : * The first line of the output should be the goal of the plan